suayptalha/Clarus-7B-v0.1
suayptalha/Clarus-7B-v0.1 is a 7 billion parameter language model created by suayptalha, merged from two Qwen2.5-7B-CABS models using the SLERP method. This model demonstrates an average performance of 36.71 on the Open LLM Leaderboard, with notable scores in IFEval (74.54) and MATH Lvl 5 (49.24). It is designed for general language tasks, leveraging the combined strengths of its base models.
Loading preview...
Model Overview
suayptalha/Clarus-7B-v0.1 is a 7 billion parameter language model developed by suayptalha. It is a merged model, combining two versions of the gz987/qwen2.5-7b-cabs base models (v0.3 and v0.4) using the SLERP merge method. This merging technique aims to combine the strengths of the constituent models to enhance overall performance.
Performance Highlights
Evaluated on the Open LLM Leaderboard, Clarus-7B-v0.1 achieved an average score of 36.71. Key performance metrics include:
- IFEval (0-Shot): 74.54
- MATH Lvl 5 (4-Shot): 49.24
- BBH (3-Shot): 36.03
- MMLU-PRO (5-shot): 37.64
These results indicate its capabilities across various reasoning and instruction-following tasks.
Merge Configuration
The model was created using mergekit with a specific YAML configuration that details the layer ranges and parameter weighting for the SLERP merge. This configuration involved blending layers from both gz987/qwen2.5-7b-cabs-v0.3 and gz987/qwen2.5-7b-cabs-v0.4 to optimize the final model's characteristics.